Object Detection And Center Localization Using MACF

With the development and the progress of science and technology, there is an increasinglyprobability to replace human beings by machines. Object detection is one of the most basicfunctions of human visual system, since Marr proposed three layers model of computer visionto understand the human visual system, more and more researchers join computer visionresearches. This paper mainly works on using computer to simulate human eyes in objectdetection.The research features object detection problem. The purpose of this work is to explore aneffective object representation method. To this end, in the following paper, we firstly studiedand discussed the previous object representation methods, and on the basis of a single scaleactive contour fragments feature, we proposed the multi-scale deformable contourinformation MACF as object features. These features were cascaded into strong detectorusing RealBoost algorithm, and this strong detector was used in object detection. Meanwhile,multi-scale deformable contour information MACF encoded the information about objectcenter position, it can vote for true object center position, different from all state-of-the artcenter localization algorithms，the proposed method considered the impact on the centerposition of the object contour segments deformation. In the locating part, using MeanShiftalgorithm，to inset the deformable information, and finally got more accurate objectlocalization result.The paper demonstrated the advantages of the proposed method on both single scalemulti-view and multi-scale multi-view car detection and localization. The experimentalresults show that our method achieves the detection performance slightly better or equivalentto that of the state-of-the-art techniques even though we have only employed less than15%of training images they used. The localization results also support the accuracy of the proposedself-calibrating object center estimation.